teamshift

teamshift

Run TeamShift agent teams from your terminal. Metered by usage — you do not bring your own model key, so a run costs you TeamShift credit rather than an Anthropic or OpenAI bill.

npx teamshift login
npx teamshift catalog                  # prebuilt teams you can add
npx teamshift add code-audit-swarm     # adds it to your workspace
npx teamshift run <team-id> "review services/api for auth mistakes"

teamshift teams lists everything callable, including teams you built yourself in the portal. add is idempotent — running it twice will not duplicate a team or overwrite edits you have made to it.

Parallel fan-out

The reason to reach for the CLI instead of the API: one command dispatches many prompts at once and reports what each one cost.

npx teamshift run <team-id> \
  -p "audit auth" -p "map the schema" -p "find dead code" --concurrency 3

STATUS     CHARGE  TIME  PROMPT
─────────  ──────  ────  ───────────────
completed  $1.20   4.1s  audit auth
completed  $1.20   4.0s  map the schema
completed  $1.20   4.0s  find dead code

3 run(s) · $3.60 total · 8.1s wall clock (12.1s if run one at a time)

--file prompts.txt reads one prompt per line. Concurrency is bounded (default 5) so a large batch does not trip the API's write rate limit.

Already using Claude Code or Codex?

Prefer the MCP server — it exposes the same teams as tools inside your existing agent, with no second CLI to learn. Both install commands are in the portal under Programmatic setup.

Auth

TEAMSHIFT_API_KEY wins over the stored config, so CI never has to write a file. teamshift login writes ~/.teamshift/config.json at mode 0600. TEAMSHIFT_API_URL overrides the endpoint.

Exit codes

Code	Meaning
0	every run reached a terminal status
1	usage error, auth failure, or at least one run failed to dispatch
2	out of credit — add credits and rerun

Releasing

This package is private: true so a workspace-wide publish cannot push it by accident. Releasing is deliberate: set private: false, pnpm build, then npm publish.